Local AI Content Workflow on iPhone — PocketPal, Shortcuts & Edits
An on-device, mobile-first pipeline for captions, transcription, AI-assisted titles, and multi-platform scheduling — no internet required*.
The Workflow at a Glance
Most content creation advice assumes you're sitting at a desk, uploading to cloud services, paying for subscriptions, and feeding your raw footage and ideas into systems you don't control or understand.
This post is about an alternative approach.
Everything described here runs directly on your phone. The AI model that generates your titles and descriptions lives on your device. The transcription happens locally. No audio leaves your phone. No account needed. No internet required after downloading the apps and their initial setup.
What You'll Need
Apps:
Edits — short-form video editor with caption generation (free)
Shortcuts — Apple's built-in automation app (pre-installed on iOS)
PocketPal AI — local LLM runner for iPhone (free)
TikTok Studio — required for TikTok scheduling (free)
YouTube Studio — for YouTube Shorts scheduling (free)
Instagram — scheduling via the native app (free)
You'll also need the Files app, which comes with every iPhone.
Part 1: Edits App As Reusable Template
The Edits app is the starting point and base video project manager. It can handle on-screencaption generation*, visual layout, and export.
I will show you how to use this app’s projects as a reusable template including some quirks and pitfalls.
*if you use on-screen captions it requires an internet connection and sends your data to Meta for audio transcription.
There's a small but important trick to keep your template intact when swapping videos.
Open the Edits app and load your existing project. Scroll to the end of the timeline and add your new video there first — don't drop it in at the top. Then delete the old video.
Here's the catch: if you delete a clip normally, it takes your headings, text overlays, and timeline positioning with it. To avoid that, go into Settings within the project and turn off Visual Clips. Now when you delete the old video, the text elements stay put.
Next, delete the old captions and tap Generate New Captions. The Edits app transcribes your audio automatically* — it's fast and reasonably accurate, but it has a quirk: it consistently mishears my name as "Andrew." If you see something similar, just tap the word to edit it inline.
Once captions are done, scroll to the end to confirm your video runs under 3 minutes. Aim for 2:59, as I’ve had videos that are exactly 3:00 in the edits app not qualify as a ‘short’ when uploaded to YouTube for being over its 3min limit.
One thing to watch: if you trimmed the video, make sure you've also trimmed the caption timeline above it — captions extending past the video end can push you over the limit.
The title gets added last, after we've run the transcript through PocketPal.
Part 2: Extracting Audio with the iOS Shortcuts App
This is where the local pipeline really starts. Instead of uploading your video to a cloud transcription service, you're going to extract the audio on-device using two custom Shortcuts you build yourself.
Why build them yourself? I'll share download links below, but I'd encourage you to build these from scratch first. Downloading Shortcuts from the internet is a genuine security consideration — a Shortcut is essentially a small automation script, and a malicious one could do things you didn't intend. Apple does sandbox them to some degree, but the best practice is to understand what a Shortcut does before you run it. Building your own takes about five minutes and means you know exactly what's happening.
Shortcut 1: Get Audio
Open the Shortcuts app (if you've deleted it, it's a free re-download from the App Store). Tap the + to create a new shortcut.
Search for Encode Media and select it — choose Audio Only
Tap Media and select Shortcut Input
When it asks for input type, select Ask for and untoggle everything except Media (this limits it to video files)
Search for Save File, add it, and expand the options
Untoggle Ask Where to Save, then tap the location field and choose a consistent folder (e.g. a folder called "Audio Exports" in your Files app)
Rename the shortcut — something like Get Audio — and change the icon if you like
Tap the ⓘ at the bottom and hit Add to Home Screen
Bonus: search ‘clipboard’ and add ‘copy to clipboard’ as a final action to have the transcript ready in your paste into PocketPAL when the transcription finishes.
Shortcut 2: Get Transcript
Search for Transcribe Audio and select it — leave it set to audio file input
Add Save File, expand, untoggle Ask Where to Save, choose the same folder as above
Rename it Get Transcript, update the icon
Add to Home Screen
To use them:Get Audio on your home screen, select your video (the same base video you added to Edits), and allow access. A plain audio file will appear in your chosen folder. Then tap Get Transcript, select that audio file (use the timestamp if the filename is unclear), allow access, and the transcription runs immediately — you'll watch the text appear in real time. The resulting .txt file saves to the same folder.
Download links (use at your own discretion — build them yourself if you're security-conscious):
Part 3: PocketPal — Your On-Device AI
PocketPal runs a small language model entirely on your iPhone. No account. No API key. No internet. Once the model is downloaded, it works offline.
Setting Up PocketPal
Download PocketPal AI from the App Store (iOS) or Google Play Store (Android).
Open the app and go to the Models tab
Search for and download Gemma-2-2B-IT (Q6_K) — this is the model I use. It's small enough to run well on an iPhone but capable enough for title and description work. The download is a few gigabytes, so do it on Wi-Fi
(Not sure what Q6_K means? It refers to the quantisation level — essentially how compressed the model is. Higher numbers mean better quality but more RAM required. Q6_K is a good balance for mobile.)
Creating a Custom PAL
A PAL in PocketPal is a saved persona — a system prompt that tells the model how to behave before you say anything. This example is what lets you paste a raw transcript and immediately get useful output without writing a long prompt every time.
Go to the PALs tab and tap +
Give it a name — something like Content Assistant or Transcript PAL
In the system prompt field, paste the following:
[PLACEHOLDER — paste your system prompt here from the previous post]
Save the PAL
When you start a new chat, select this PAL from the dropdown before sending anything
Running the Transcript
Go to the Files app, find your .txt transcript, tap and hold, and select Copy. Open PocketPal, start a new chat with your PAL selected, paste the transcript, and send.
The model will generate title options and a description draft. Watch it type in real time — on an iPhone 15 Pro or similar, it moves fast. This is genuinely happening on your device.
Once it's done, hold down on the output and copy it. Paste it into the Notes app immediately — the clipboard can clear itself, and you don't want to lose it.
From there, spend a minute or two in Notes cleaning up: pick the title you like, adjust any emoji, tighten the description. Then you're ready to publish.
Part 4: Scheduling Across Platforms
Final Edits App Procedure (and Instagram Scheduling)
Go back to the Edits app, open your project, tap the title field and paste in your new title
Export the video — this takes a while and requires the screen to stay on (see below)
When export is complete, tap Download to Gallery so it can be used later by the other apps)
Click the Instagram logo to share the post directly to Instagram from the Edits app
Select your reel, go to More Options → Schedule This Reel, pick your date and time
You must keep the screen on and not change apps while it's uploading
Important: each app requires the screen to stay on while uploading, do not switch to another app. Go to Settings → Display & Brightness → Auto-Lock and set it to Never. Plug in while it uploads. Change it back afterwards.
To check scheduled posts: tap the top-right menu in Instagram → Scheduled Content.
TikTok
You need the TikTok Studio app specifically — you can't schedule from the main TikTok app. The standard TikTok app doesn't support scheduling on mobile; Studio does. (You can also schedule from TikTok's web interface on desktop.)
Open TikTok Studio, tap +, find your exported video, tap Next
Paste in your description, adjust the cover if needed
Tap Schedule Post and choose your date and time
Hit Post — the scheduled video will appear at the top of your posts list
You must keep the screen on and not change apps while it's uploading
YouTube Shorts
Open YouTube Studio, tap +
Select your exported video
Tap Show More, paste your description first, then paste/type your title
Adjust any other settings, tap visibility and select Schedule, set your date and time
Tap Upload Short
YouTube upload speed depends on your connection. You must keep the screen on and not change apps while it's uploading.
Why This Matters
Most content creation pipelines are designed to make you dependent: upload here, transcribe there, generate titles with this subscription, schedule through that platform's paid tier. Your raw footage, your voice, your ideas — all of it passing through servers you don't control, feeding models you didn't consent to train.
The local-first approach isn't a complete solution to that. You're still publishing to centralised platforms at the end. But the production layer — the most sensitive part, where your unedited, unpolished creative work lives — stays on your device.
Small models like Gemma 2B aren't as capable as GPT-4 or Claude for complex reasoning. They'll occasionally produce clunky titles or miss the tone. That's a real trade-off and it's worth being honest about. But for a focused task like "read this transcript and give me five title options and a short description," they do the job — and they do it without an internet connection, a subscription, or a data transfer.
Quick Reference: The Full Flow
Edits app — swap video, generate captions, check timing
Get Audio shortcut — extract audio from exported video
Get Transcript shortcut — transcribe audio to text file
PocketPal — paste transcript, generate titles + description with your custom PAL
Notes app — clean up and store output
Edits app — add final title, export
Instagram / TikTok Studio / YouTube Studio — schedule
Total active time once you have the workflow set up: roughly 15–20 minutes per video.
Related Posts
Run Your Own ChatGPT on a Computer at Home — the desktop version of this local AI approach
